How it works

  1. Configure. Select a deployment template and state GPU family and count, minimum VRAM per GPU, CPU/RAM needs, disk, region, rental window, expected transfers and a budget cap. Billing units are shown before any comparison.
  2. Compare. Only eligible offers are ranked — by total estimate. Region, memory, group sizes, availability and budget failures are explained, never silently bypassed.
  3. Review. The accepted quote pins offer, template, rate-card version, quantities and terms. Funding is confirmed before activation; the cap is a ceiling, not a charge.
  4. Price and surge. Under pricing model v0.2 the price is tokens + GPU time + demand surge k (0.7–2.0). Price and k are shown before confirmation, the quote is locked for 10 minutes, and a booking keeps its k for the full term.
  5. Compliance. KYC/KYB, sanctions screening (OFAC, EU, UK, UN, Armenia), declaration of purpose.
  6. Instance. Provisioning, running, stopped, offline and released states each have defined billing: running accrues GPU + disk, stopped accrues retained disk only, released stops everything.
  7. Statement. Metered usage reconciles to accepted rates with the seller payout (80% of base + 60% of surge, never below 65% of base), the included platform share and any unused return — all from retained records.

T-Exchange coordinates ordering, metering, pricing and settlement from Yerevan, Armenia. Sellers run the instances and stand behind their terms. Software templates stay separate from resource offers throughout.

Ask